Association of Sequential Organ Failure Assessment (SOFA) components with mortality

Abstract Background Sequential Organ Failure Assessment (SOFA) is a practical method to describe and quantify the presence and severity of organ system dysfunctions and failures. Some proposals suggest that SOFA could be employed as an endpoint in trials. To justify this, all SOFA component scores should reflect organ dysfunctions of comparable severity. We aimed to investigate whether the associations of different SOFA components with in‐hospital mortality are comparable. Methods We performed a study based on nationwide register data on adult patients admitted to 26 Finnish intensive care units (ICUs) during 2012−2015. We determined the SOFA score as the maximum score in the first 24 hours after ICU admission. We defined organ failure (OF) as an organ‐specific SOFA score of three or higher. We evaluated the association of different SOFA component scores with mortality. Results Our study population comprised 63,756 ICU patients. Overall hospital mortality was 10.7%. In‐hospital mortality was 22.5% for patients with respiratory failure, 34.8% for those with coagulation failure, 40.1% for those with hepatic failure, 14.9% for those with cardiovascular failure, 26.9% for those with neurologic failure and 34.6% for the patients with renal failure. Among patients with comparable total SOFA scores, the risk of death was lower in patients with cardiovascular OF compared with patients with other OFs. Conclusions All SOFA components are associated with mortality, but their weights are not comparable. High scores of other organ systems mean a higher risk of death than high cardiovascular scores. The scoring of cardiovascular dysfunction needs to be updated.


| INTRODUC TI ON
The Sequential Organ Failure Assessment (SOFA), at first named Sepsis-related Organ Failure Assessment, was introduced by The Working Group on Sepsis-Related Problems in 1996. 1 The SOFA score describes and quantifies the severity of dysfunction or failure of six essential organ systems (Table S1). Primarily, the SOFA score was not meant for outcome prediction. Multiple studies have shown, however, that it can rather well predict mortality in groups of critically ill patients. [1][2][3][4][5][6][7][8][9][10][11][12][13] This has notably widened the employment of the SOFA score beyond its original purpose.
In randomised controlled trials, the gold standard has been to use all-cause mortality as an endpoint. However, interventional trials often fail to detect any difference between study arms in mortality. 14 Therefore, there is growing interest to use surrogate endpoints, for example SOFA scores. [15][16][17] Regulatory authorities, including the European Medicines Agency, can under certain limitations approve the use of surrogate endpoints instead of mortality as primary endpoints. 18 The change in the SOFA score during critical illness has been proposed to reflect the benefit or harm of the intervention of interest.
The SOFA score, which is a scalar variable, is presumably more sensitive in detecting the effects of an intervention than mortality, a binary variable. However, the total SOFA score cannot be an unbiased trial endpoint unless all its components have comparable weights as measures of organ dysfunction severity. Moreover, some organ failures (OFs) are more likely to occur concurrently. 3 It is unclear whether different combinations of OFs affect the predictive value of total SOFA score.
The aim of this study was to investigate whether different SOFA score components, recorded during the first 24 h of intensive care, carry comparable weights in terms of their association with mortality. In other words, do patients with comparable total SOFA scores have comparable probabilities to perish regardless of which OFs they suffer from? We evaluated how combinations of different organ system failures are associated with mortality. Furthermore, we assessed the association of increasing SOFA scores with mortality across different admission groups. Mortality at hospital discharge was the primary endpoint. Mortality at ICU discharge and mortality within 12 months were secondary endpoints. We included all adult patients admitted to Finnish ICUs between January 1, 2012 and December 31, 2015. For patients with multiple ICU treatment periods during the same hospitalisation, we included only the first ICU admission. In line with the 1998 paper by the working group that created the SOFA system, 2 we defined OF as an organ-specific SOFA score of three or higher. OF could appear isolated or as part of multiorgan failure.

| Study design and participants
We performed subgroup analyses to observe whether the findings were consistent, regardless of the admission type-medical, elective surgery and emergency surgery.

| Extracted variables
We extracted following variables from the FICC database: the most severe values of SOFA score components within the first 24 h after admission to the ICU and the outcome variables: vital status at ICU discharge, at hospital discharge and 12 months after ICU admission. Moreover, we gathered baseline data on Acute Physiology, Age, Chronic Health Evaluation (APACHE) II, 20 The Simplified Acute Physiology Score (SAPS) II, 21 age and sex. We also retrieved data on length of stay in the ICU and length of stay in hospital.

| Data handling and statistical methods
In the neurologic component, the SOFA score is based on the Glasgow Coma Score (GCS). For anaesthetised or sedated patients, the GCS recorded to the FICC registry is the last reliable GCS preceding sedation, in line with the SAPS II score. 21 The hepatic SOFA score is based on the plasma bilirubin concentration. Bilirubin is normally measured when there is a clinical reason to suspect hepatic problems. Therefore, we consider normality of bilirubin concentrations as likely in patients for whom the data on bilirubin were missing. In these patients, we assumed the hepatic SOFA score to be 0. We made no assumption of normality for other SOFA components in cases of missing data. Therefore, we excluded patients with missing SOFA data concerning all other components except for the hepatic component. In addition, we excluded patients with missing mortality data.
We compared the characteristics of survivors and non-survivors at hospital discharge employing the Mann−Whitney U-test for continuous data and chi square test for categorical data. Using age-adjusted multivariable logistic regression, we evaluated the association between SOFA score components and mortality. All components as well as age were included in the analysis simultaneously. P-value of less than 0.05 was considered as statistically significant in all tests. We calculated standardized occurrence ratio (SOR) for each set of at least two, three, or four concurrently occurring failing organ systems. SOR is a tool to evaluate whether particular OFs occur concurrently more frequently than anticipated by merely observing the frequencies of OFs. SOR was calculated as where N(o) is the number of patients with OF of a and b, N is the total number of admissions and p(a) and p(b) are the proportions of patients with failure of organ systems a andb, respectively. In the same way, we calculated the SOR for patients with three and four concurrent OFs. SOR >1 signals that the odds of concurrent occurrence of these particular failing organ systems are increased. Bonferroni correction was used for multiple comparisons regarding the SOR analysis.

| Study population
There were totally 71,492 ICU admissions during the study period.  Table 1. The median score in the respiratory component was 2 and in the cardiovascular component 3. In the cardiovascular component, the scores were almost equally distributed among the patients except for score 2, which was documented for only 678 (1.1%) patients. In all other components (coagulation, hepatic, neurological and renal), the median score was 0, with the score 1 being second most common ( Figure 2). Of OFs, defined as an organ-specific SOFA score ≥3, the most common OF was cardiovascular failure, in 53.6% of patients.
The second most common OF was respiratory failure, in 22.5% of patients.

| ICU, hospital and 12-month mortality
Overall, 6,851 (10.7%) patients died in hospital. The first day total SOFA score was strongly associated with mortality ( Figure 3).
Mortality was 5.3% at ICU discharge and 21.6% in 12 months.
Mortality increased with increasing SOFA scores ( Figure 3). Inhospital mortality was 15.0% in those patients with LOS at the ICU more than 48 h. There were 642 (1%) patients with a SOFA score over 15. In these patients, ICU mortality was 60%, hospital mortality was 72%, and 12-month mortality was 80%. This pattern appeared rather similar regardless of whether the vital status was observed at ICU or hospital discharge or 12 months after ICU admission (Figures 4 and 5).
Mortality in patients with OFs (organ-specific SOFA score 3 or 4) increased with increasing total SOFA scores. However, within groups of patients with comparable total SOFA scores, mortality was lower in patients with cardiovascular OF compared with patients with other OFs in patients with a total SOFA score lower than 12.
In fact, mortality in patients with cardiovascular OF did not exceed the mortality in patients with no first-day OF at all in patients with a total SOFA score lower than 9 ( Figure 6).

| Combinations of organ system failures and mortality
Mortality increased with increasing numbers of concurrent OFs ( Figure 7). Of all patients, 47.4% had at least two, 12.7% had at least

F I G U R E 3
The number of patients and mortality according to first-day total SOFA score. ICU mortality (blue line), in-hospital mortality (red line) and 12-month mortality (green line) increased with increasing total SOFA score. The bars present the number of patients within each total SOFA score group three and 2.0% had at least four concurrent OFs. In-hospital mortalities in these groups were 35 with first-day SOFA score above 15 was 72% in our study. The overall ICU and in-hospital mortality was lower in our study compared with that reported in the 1990s. 2,3 In addition to assumed improvements in prognosis of ICU patients, a plausible explanation for this is that we also included patients with preceding scheduled surgery.
Our results show, however, that the cardiovascular SOFA score was associated with lower risk of mortality in the whole cohort, in both emergency and elective admissions, as well as those with at least 48 h length of ICU stay.
Although high SOFA scores often indicate a poor prognosis, cardiovascular scores seem to be an exception. This may reflect a change in clinical practices in recent years. The SOFA score was introduced in an era of more restricted use of vasopressors. During the last two decades, the use of norepinephrine has become more common. [29][30][31] Vasopressor treatment is initiated earlier without preceding large doses of resuscitation fluids. [32][33][34][35] An infusion of norepinephrine lasting at least one hour, even at a small dose, assigns 3 points to the F I G U R E 6 Mortality in patients with different organ failures according to total SOFA score. The lines represent in-hospital mortality in patients with respiratory (light blue), coagulation (orange), hepatic (grey), cardiovascular (yellow), neurologic (purple) and renal (green) failure.
The organ failure was determined as organ-specific SOFA score 3 or 4. Mortality in patients without any first-day organ failure is shown with black dashed line  OFs by means of principal components analysis. 3 The group identified two common OF combinations. The first combination comprised respiratory, cardiovascular and neurologic OFs, whereas the second comprised coagulation, hepatic and renal OFs. In our study, this first combination of respiratory, cardiovascular and neurologic OFs was also the most common of the combinations with three OFs, affecting 37% of patients with at least three concurrently failing organ systems.
The in-hospital and 12-month mortalities associated with this particular combination were 41% and 55%, respectively, whereas in-hospital and 12-month mortalities of patients with other triple OF combinations ranged between 56%-82% and 63%-88%, respectively.
We found that the second combination, which comprised coagulation, hepatic and renal OFs, occurred 44 times more often than one would have expected by observing merely the frequency of these OFs in the whole study population.
There is growing interest in using the SOFA score as a surrogate endpoint for mortality in clinical trials. 17 Our findings suggest that this may not be without problems.

| Strengths and limitations of the study
Our study population consisted of a large unselected group of patients treated in Finnish ICUs. The data were retrieved from a highquality national database with all Finnish general ICUs participating.
Therefore, our study population is well representative of adult ICU patients in Finland. We do not know whether the results are generalizable to other countries. However, early use of norepinephrine has become more common in other countries as well, 31 and it is likely that the relation between cardiovascular SOFA scores and mortality may have weakened also in other countries.
A major limitation of our study is that the SOFA scores were based only on measurements during the first 24 h after admission to the ICU, whereas previous studies have shown that a change in SOFA score over time is the most reliable predictor of mortality. 38

ACK N OWLED G EM ENTS
We are grateful to the doctors and nurses in the ICUs participating in the Finnish Intensive Care Consortium for careful data documentation and to TietoEVRY for managing the database.

CO N FLI C T S O F I NTE R E S T
None.

AUTH O R CO NTR I B UTI O N S
MR presented the first idea of the study. AP analysed and inter-